Duck Creek Buys Send: The Build-vs-Buy Shift in Agentic Underwriting

Duck Creek acquired underwriting orchestration vendor Send Technology Solutions on July 7, 2026, folding a platform that already supports $26 billion in gross written premium and claims up to 7x faster time-to-quote (Duck Creek, July 2026) into its five-month-old Agentic AI Platform, making orchestration, not the scoring model underneath it, the layer where core-system vendors are now consolidating.

What Send's Orchestration Engine Actually Automates

Send is not a predictive score or a rating engine. It is a workflow layer that sits between the moment a submission arrives and the moment an underwriter makes a decision, and its scope is broader than the "triage" label usually applied to it. The platform ingests submissions from email, documents, and broker channels, structures the unstructured data, flags missing fields, and coordinates the workflow through risk assessment, pricing, approvals, quote, bind, and post-bind stages, all inside what Duck Creek describes as a risk-centered workspace consolidating documents, data, notes, and decisions in one place (Duck Creek, July 2026). It was purpose-built for commercial, specialty, MGA, delegated authority, reinsurance, and London Market business, the segments where submission complexity, multi-section programmes, and subscription placements make manual data assembly the biggest drag on underwriter throughput.

The human underwriter has not been removed from that workflow, but their position in it has shifted upstream. Send's configurable rules determine which submissions get auto-quoted within appetite, which get escalated for manual review, and which get auto-declined before a human ever opens the file. "Send was built to help insurers navigate increasingly complex underwriting decisions by orchestrating people, data, AI, and workflows into a single underwriting experience," said Andy Moss, Send's chief executive officer and co-founder, who will continue leading the Send team at Duck Creek as general manager of underwriting (Duck Creek, July 2026). Moss added that "technology augments underwriters, not replaces them," but the practical effect of an orchestration layer is that the underwriter's judgment now applies to a pre-filtered subset of submissions rather than the full inbound flow. That is precisely why the rules governing the filter matter as much as the rules governing the eventual decision.

The Second-Order Problem With 7x Faster and 65% Shorter

Duck Creek and Send are marketing two headline metrics: up to 7x faster time-to-quote and up to a 65% reduction in product launch cycles (Duck Creek, July 2026). Both are throughput measurements, and neither one, on its own, says anything about underwriting quality. A system quotes faster when it processes more submissions automatically within a defined rule set. It also quotes faster when it declines or routes away more submissions that fall outside that rule set before a human ever sees them. Those are two different mechanisms producing the same headline number, and only one of them represents a genuine efficiency gain rather than a shift in what gets measured.

This is the adverse selection question that trade coverage of the deal has skipped entirely. If Send's triage rules systematically under-price or over-quote a particular class, territory, or risk profile relative to the carrier's actual loss experience, that bias will not show up as a validation failure at deployment. It will show up eighteen to twenty-four months later as an unexplained deterioration in loss ratio for the segments the orchestration engine was quoting fastest, by which point a full accident year of business has already been written on rules nobody outside the vendor's own team independently re-tested. The fix is not complicated in principle: carriers need loss-ratio-by-disposition reporting that separates auto-quoted, escalated, and auto-declined submissions, tracked against the rule version in force at the time each policy was written. Whether Send's current customer base is producing that reporting today is an open question the acquisition does not answer, and it is the first thing a pricing actuary evaluating this platform should ask for before treating the 7x figure as a proxy for underwriting soundness.

Two Deals, Two Different Layers of the Stack

Duck Creek's acquisition of Send is the second major insurance-AI deal in six weeks to buy a specific rung of the AI stack rather than build it internally, and the two deals bought different rungs. EXL announced on June 24, 2026 that it would acquire iMerit, an AI model training and annotation firm, for $310 million: $170 million upfront plus $140 million in earnouts contingent on performance milestones over two years (ExlService Holdings, June 2026). iMerit's assets, the Ango Hub annotation platform and the Scholars network of domain experts, sit at the training and evaluation layer of the AI stack, the work that happens before a model is deployed. Send sits downstream of that, at the orchestration and decisioning layer, the work that happens after a model exists and needs to be wired into a live underwriting workflow.

What makes the Duck Creek deal more notable is that Duck Creek had already built its own orchestration layer organically. The company launched its insurance-native Agentic AI Platform at its Formation '26 conference on April 28, 2026, introducing a five-layer architecture that included an Agentic Orchestration component and an Agentic Underwriting Workbench application, built around what it calls a Model Context Repository combining fine-tuned generative models with neuro-symbolic reasoning grounded in carrier-specific rules. Ten weeks later, rather than extend that homegrown orchestration workbench, Duck Creek bought a specialist vendor with $26 billion in GWP already flowing through its rules and years of London Market and delegated authority workflow depth. Hardeep Gulati, Duck Creek's chief executive officer, framed the logic directly: "By bringing Send into Duck Creek, we are combining core insurance operations and underwriting orchestration in a trusted agentic AI experience that connects systems of record, decision intelligence, workflows, and action" (Duck Creek, July 2026). Read against the April launch, that is an admission that building orchestration logic from scratch, even with a dedicated platform effort, moves slower than buying a vendor whose triage rules are already proven against $26 billion of live premium.

Where the Underwriting Guidelines Now Live

Orchestration engines do not just route paperwork. They encode underwriting guidelines, appetite thresholds, and auto-quote conditions as executable rules, which means the actual content of a carrier's underwriting manual increasingly lives as configuration inside a third party's software rather than as an internal document the carrier's own underwriting and compliance teams fully control. Send's platform explicitly markets "configurable workflows, rules, governance, and compliance controls" as a core feature, spanning more than 40 lines of business (Send Technology Solutions, 2026). That configurability is the product. It is also the governance exposure. Most states require insurers to be able to produce the underwriting criteria supporting a rate filing on request, and triage or auto-decline logic that systematically shapes which risks reach a pricing engine functions as underwriting criteria even when it is not itself a rating factor. The NAIC's Model Bulletin on the Use of Artificial Intelligence Systems by Insurers, adopted in December 2023 and in force in close to half of U.S. states as of 2026, requires insurers to maintain written standards for the acquisition and use of third-party AI systems, to conduct due diligence on vendor data and models, and to secure contractual audit rights and cooperation with regulatory inquiries (NAIC, 2024). A carrier that cannot produce Send's current triage rule set, the version history behind it, and the loss experience validating it, on the same timeline it would produce its own internal underwriting manual, has not actually met that standard just because the vendor's platform carries built-in governance tooling. The NAIC's Third-Party Data and Models (H) Task Force, formed in 2024 specifically to build a regulatory framework around vendor-supplied AI, is the body most likely to test that gap first.

The Total Cost of Ownership When the Product Is Orchestration

Build-versus-buy analysis for actuarial AI has historically centered on a single predictive model: build a proprietary GLM or gradient-boosted score in-house, or buy a vendor score and integrate it through an API. The total cost of ownership math for that decision is relatively contained, dominated by data acquisition cost, one-time integration effort, and periodic model refresh. Buying an orchestration engine changes the shape of that math in three ways that carriers evaluating a Send-like platform, whether through Duck Creek or a competitor, need to price explicitly.

First, orchestration is not a model validated once at deployment; it is a live rules layer reconfigured continuously as new products launch, which is exactly what the 65% product-launch-cycle claim describes. Every reconfiguration is either a carrier-side change request routed through the vendor or a vendor-side change the carrier must re-validate, and that ongoing change-management cost recurs for the life of the contract in a way a static scoring API's refresh cycle does not. Second, the audit and documentation burden shifts from a one-time model validation exercise to continuous proof that the vendor's live rule set still matches what was represented in the last rate filing or market conduct exam, which is a standing compliance cost rather than a sunk one. Third, and most consequential for a multi-year contract, switching costs are far higher than for a scoring vendor. Migrating away from a proprietary orchestration engine means re-encoding potentially thousands of triage and appetite rules into a new vendor's rule format, not swapping one API endpoint for another, which locks in negotiating leverage on the vendor's side for the duration of the relationship. Carriers with mature in-house rules-authoring tools, the kind Guidewire and Duck Creek both already sell as part of their core policy administration suites, have the weakest case for buying a standalone orchestration layer; carriers without that infrastructure, or racing to match a competitor's quote speed, have the strongest case and the least leverage once they sign.

Guidewire, Sapiens, and Socotra Draw the Line Differently

Duck Creek's acquisition also throws into relief how differently its three main competitors are handling the same orchestration-versus-scoring boundary. Guidewire has built an Agentic Framework that lets developers build, test, and deploy custom AI agents on top of Guidewire Cloud APIs, or integrate agents from partner vendors, but it has not acquired a standalone orchestration specialist the way Duck Creek just did. Sapiens has taken a similar partner-integration posture rather than an acquisition posture. Socotra, built natively on open APIs and public cloud infrastructure, leaves orchestration to whatever a carrier or MGA chooses to build or plug in, positioning itself as infrastructure rather than an opinionated workflow layer.

The clearest evidence of the gap Duck Creek just closed for itself came from a third party. Convr, an AI underwriting workbench vendor, announced on the same day as the Duck Creek deal that it delivers AI-powered intake, enrichment, classification, scoring, and agentic decisioning across Guidewire, Duck Creek, and Sapiens alike, describing itself as "the connective tissue between underwriting intelligence and core systems, not a replacement for either," in the words of chief executive officer John Stammen (Convr, July 2026). That framing captures the current state of the market precisely: Guidewire and Sapiens carriers still depend on a shared, core-system-agnostic third party for the orchestration layer, while Duck Creek, as of this acquisition, now owns a proprietary version of that same layer outright. Whether owning it outright is a durable advantage depends on whether Duck Creek can integrate Send's rule engine into its Model Context Repository faster than Guidewire and Sapiens can either acquire a comparable specialist themselves or deepen their existing partnerships with vendors like Convr.

Why This Matters

The Duck Creek-Send deal is being read in trade press as a product bundling story: one more vendor claiming an end-to-end platform. For actuaries, the more consequential read is that the physical location of underwriting guideline logic is consolidating into a small number of orchestration vendors faster than the governance frameworks meant to oversee it are maturing. Pricing and underwriting actuaries evaluating any orchestration platform, Send's or a competitor's, should treat the throughput metrics vendors publish as a starting hypothesis rather than a procurement decision, and should demand disposition-level loss data, segmented by auto-quoted, escalated, and auto-declined status and tied to specific rule versions, before renewing or expanding a contract. Reserving and model governance teams should confirm, in writing, that they can produce a vendor's current triage rule set on the same timeline as an internal underwriting manual, because that is the standard the NAIC's Third-Party Data and Models Task Force is moving toward, not a hypothetical one. The next wave of consolidation in this market will not be won by whichever vendor claims the fastest quote time. It will be won by whichever vendor can prove, on demand, that its rules were fair to begin with.